High-speed bearing fault diagnosis method and equipment based on coupling model and adversarial network

Abstract

The invention provides a high-speed bearing fault diagnosis method and equipment based on a coupling model and an adversarial network, and belongs to the technical field of bearing fault diagnosis, and the method comprises the steps: carrying out the dynamics simulation based on an axle box-vehicle-track coupling dynamics model, obtaining a source domain, and obtaining a target domain, the target domain is a bearing data set which is not marked with a label; performing data preprocessing on the source domain and the target domain; pre-training the constructed domain adversarial neural network based on the preprocessed source domain until the maximum number of iterations is reached, and obtaining a pre-trained domain adversarial neural network; inputting the pre-processed target domain into a pre-trained domain adversarial neural network for adjustment so as to identify the health state of the bearing; the source domain and the target domain cover bearing faults of more working conditions, the difference between the source domain and the target domain is reduced, meanwhile, fault features of multiple scales are extracted through the multi-scale expansion convolution layer, and the diagnosis accuracy of the bearing faults is improved.

Description

Technical Field

[0001] This application belongs to the field of bearing fault diagnosis technology, and in particular relates to a high-speed bearing fault diagnosis method and device based on coupled models and adversarial networks. Background Technology

[0002] Currently, high-speed train bearing fault diagnosis uses traditionally trained deep learning neural networks. However, deep learning neural networks require a large number of labeled samples and training and testing data with uniform data distribution to improve the accuracy of bearing fault diagnosis. Currently, high-speed trains suffer from insufficient real-vehicle bearing fault data, resulting in unsatisfactory diagnostic accuracy.

[0003] Existing technologies use experimental benches to generate bearing fault data for various operating conditions. However, the source domain bearing fault data differs significantly from the target domain bearing data and cannot cover more operating conditions. Some abnormal source domain bearing fault data reduces the accuracy of bearing fault diagnosis.

[0004] Existing bearing fault diagnosis technologies suffer from insufficient data on the types of operating conditions for bearing faults, and the presence of abnormal source domain bearing fault data reduces the accuracy of bearing fault diagnosis. Summary of the Invention

[0005] This application provides a high-speed bearing fault diagnosis method, electronic device, and storage medium based on a coupled model and adversarial network. It can solve the problems of insufficient operating condition types of bearing fault data and abnormal source domain bearing fault data reducing the accuracy of bearing fault diagnosis.

[0006] In a first aspect, embodiments of this application provide a high-speed bearing fault diagnosis method based on a coupled model and adversarial networks, including:

[0007] Dynamic simulations were performed based on the axle box-vehicle-track coupled dynamics model to obtain the source domain and the target domain. The source domain is a labeled bearing fault dataset with multiple working conditions, and the target domain is an unlabeled bearing dataset.

[0008] Data preprocessing is performed on the source domain and the target domain to obtain the preprocessed source domain and the preprocessed target domain;

[0009] The pre-trained domain adversarial neural network is pre-trained based on the pre-processed source domain until the maximum number of iterations is reached, resulting in a pre-trained domain adversarial neural network.

[0010] The pre-processed target domain is input into a pre-trained domain adversarial neural network for adjustment in order to identify the health status of the bearing;

[0011] Among them, the parameters for improving domain discrimination in the domain adversarial neural network include source domain sample weights, which are used to reduce the interference of abnormal bearing fault data in the source domain.

[0012] The convolutional layers of the domain adversarial neural network include multi-scale dilated convolutional layers, which are used to extract bearing fault features at multiple scales.

[0013] In one embodiment, the domain adversarial neural network includes an input signal, a feature extractor, a domain discriminator, a feature classifier, and a loss decision output;

[0014] Among them, the feature extractor is used to extract bearing fault features from the preprocessed source domain or preprocessed target domain of the input signal using a multi-scale expansion method;

[0015] The input of the domain discriminator is connected to the output of the feature extractor to receive bearing fault features. The source domain sample weights are set according to the difficulty of the domain discriminator in the parameters of the improved domain discriminator, and the difference between the bearing fault features corresponding to the source domain and the target domain is confused. The bearing fault features are labeled with the source domain or the target domain through feature input, gradient inversion layer and domain discriminator.

[0016] The input of the feature classifier is connected to the output of the feature extractor, and is used to receive bearing fault features and perform label classification on the bearing fault features.

[0017] The loss determination output receives the label to determine the classification loss, receives the domain label to determine the domain discrimination loss, and determines the joint maximum average difference loss based on the feature extraction and feature classification of the source and target domains. The marginal distribution and conditional distribution between the source and target domains are jointly aligned, and the bearing health status of the bearing fault data is identified by minimizing the loss function value.

[0018] In one embodiment, the domain adversarial neural network includes 5 convolutional layers, 1 multi-scale dilated convolutional layer, 1 adaptive max pooling layer, and 2 fully connected layers.

[0019] The feature extractor consists of five convolutional layers and one multi-scale dilated convolutional layer. Each convolutional layer consists of a convolution operation, batch normalization, and a ReLU activation function. The multi-scale dilated convolutional layer is located between the first and second convolutional layers. The second, third, fourth, and fifth convolutional layers are connected sequentially. The neighborhood discriminator includes feature embedding, a gradient inversion layer, and neighborhood discrimination. The feature classifier includes two fully connected layers.

[0020] In one embodiment, the pre-trained domain adversarial neural network is pre-trained based on the pre-processed source domain until the maximum number of iterations is reached, resulting in a pre-trained domain adversarial neural network, including:

[0021] Initialize the network structure of the domain adversarial neural network, and determine the maximum number of iterations and the learning rate;

[0022] Based on the first parameter, second parameter, third parameter, feature classifier, feature extractor, preprocessed source domain, number of samples in the source domain, number of health categories, and indicator function, the classification loss is determined by the classification loss calculation formula.

[0023] Based on the preprocessed source domain, preprocessed target domain, number of samples in the source domain, number of samples in the target domain, Gaussian kernel function, feature classifier, and feature extractor, the joint maximum average difference loss is determined by the joint maximum average difference loss calculation formula.

[0024] Based on the preprocessed source domain, preprocessed target domain, number of samples in the source domain, number of samples in the target domain, source domain sample weights, feature classifier, feature extractor, and domain discriminator, the domain discrimination loss based on the source domain sample weights is determined by the domain discrimination loss calculation formula.

[0025] The loss function value is determined by the loss function calculation formula based on classification loss, domain discrimination loss, regularization coefficient, joint maximum average difference loss and trade-off parameters.

[0026] The first, second, and third parameters are updated using an adaptive motion estimation method and a learning rate, and then iterated repeatedly until the maximum number of iterations is reached, thus obtaining a pre-trained neural network.

[0027] In one embodiment, the loss function is calculated as follows:

[0028]

[0029] Where, L(θ) f ,θ c ,θ d ) represents the loss function value; θ f θ is the first parameter of the feature extractor. c θ is the second parameter of the feature classifier. d This is the third parameter of the domain discriminator;

[0030] L c The loss is the classification loss; λ is the regularization coefficient. Loss is determined by the domain;

[0031] μ is the trade-off parameter; L JMMD This represents the combined maximum average difference loss.

[0032] In one embodiment, the classification loss is calculated as follows:

[0033]

[0034] Among them, L c For classification loss; θ f This is the first parameter of the feature extractor;

[0035] θ c This is the second parameter of the feature classifier;

[0036] n s The number of samples in the source domain;

[0037] K s The number of health categories; I[*] is the indicator function;

[0038] i is the source domain sample number; This represents the i-th bearing fault data in the source domain;

[0039] for Corresponding real-world operating condition categories;

[0040] For feature classifier G c right The k-th predicted working condition category is output.

[0041] For feature extractor G f right Output bearing fault characteristics.

[0042] In one embodiment, the domain discrimination loss is calculated as follows:

[0043]

[0044] in, For domain discrimination loss; θ f This is the first parameter of the feature extractor;

[0045] θ d This is the third parameter of the domain discriminator;

[0046] n s n represents the number of samples in the source domain. t The number of samples in the target domain;

[0047] i represents the source domain sample number; j represents the target domain sample number;

[0048] Let be the source domain sample weight of the i-th sample in the source domain;

[0049] This represents the i-th bearing fault data in the source domain; This refers to the i-th bearing data in the target domain.

[0050] For feature extractor G f right Output bearing fault characteristics;

[0051] For the domain discriminator G d right The output prediction domain category;

[0052] For feature classifier G c right Output predicted operating condition categories;

[0053] For the domain discriminator G d right The output prediction domain category;

[0054]

[0055] In one embodiment, the formula for calculating the joint maximum average difference loss is:

[0056]

[0057] Among them, L JMMD (θ f ,θ c ) represents the joint maximum average difference loss;

[0058] n s n represents the number of samples in the source domain. t The number of samples in the target domain;

[0059] i represents the source domain sample number; j represents the target domain sample number;

[0060] g(·) is the Gaussian kernel function;

[0061]

[0062] This represents the i-th bearing fault data in the source domain; This refers to the i-th bearing data in the target domain.

[0063] G c For feature classifiers; G f For feature extractors.

[0064] Secondly, embodiments of this application provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the high-speed bearing fault diagnosis method based on a coupled model and adversarial network as described in any one of the first aspects above.

[0065] Thirdly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the high-speed bearing fault diagnosis method based on a coupled model and adversarial network as described in any one of the first aspects above.

[0066] It is understood that the beneficial effects of the second and third aspects mentioned above can be found in the relevant descriptions in the first aspect above, and will not be repeated here.

[0067] The beneficial effects of the embodiments in this application compared with the prior art are:

[0068] This application provides a high-speed bearing fault diagnosis method based on a coupled model and adversarial networks. It obtains a source domain and a target domain through dynamic simulation based on an axle box-vehicle-track coupled dynamic model. The source domain consists of labeled bearing fault datasets under multiple operating conditions, while the target domain consists of unlabeled bearing datasets. Data preprocessing is performed on the source and target domains to obtain preprocessed source and target domains. A pre-built domain adversarial neural network is pre-trained based on the preprocessed source domain until the maximum number of iterations is reached, resulting in a pre-trained domain adversarial neural network. The preprocessed target domain is then input into the pre-trained domain adversarial neural network for adjustment to identify bearing health status. Compared with existing technologies, the source and target domains obtained by dynamic simulation in this application cover bearing faults under more operating conditions, reducing the difference between the source and target domains and lowering the data requirements of the domain adversarial neural network for source domain training. Furthermore, the parameters for improving domain discrimination in the pre-trained domain adversarial neural network include source domain sample weights, reducing interference from abnormal bearing fault data in the source domain. Simultaneously, multi-scale dilated convolutional layers extract bearing fault features at multiple scales, thereby improving the accuracy of bearing fault diagnosis. Attached Figure Description

[0069] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0070] Figure 1 This is a flowchart illustrating a high-speed bearing fault diagnosis method based on a coupled model and adversarial network provided in an embodiment of this application.

[0071] Figure 2 This is a schematic diagram of the framework of a domain adversarial neural network provided in an embodiment of this application;

[0072] Figure 3 This is a schematic diagram of the network structure of a domain adversarial neural network provided in an embodiment of this application;

[0073] Figure 4 This is a flowchart illustrating step S3 of an embodiment of the present application, which involves pre-training the constructed domain adversarial neural network based on the pre-processed source domain until the maximum number of iterations is reached, to obtain the pre-trained domain adversarial neural network.

[0074] Figure 5 This is a schematic diagram of a source domain provided in an embodiment of this application;

[0075] Figure 6 This is a schematic diagram of the target domain provided in an embodiment of this application;

[0076] Figure 7 This is a schematic diagram of the identified bearing health status provided in an embodiment of this application;

[0077] Figure 8 This is a confusion matrix diagram of the identification results provided in an embodiment of this application;

[0078] Figure 9 This is a t-SNE clustering diagram provided in one embodiment of this application. Detailed Implementation

[0079] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0080] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0081] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0082] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0083] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0084] Currently, high-speed train bearing fault diagnosis uses traditionally trained deep learning neural networks. However, deep learning neural networks require a large number of labeled samples and training and testing data with uniform data distribution to improve the accuracy of bearing fault diagnosis. Currently, high-speed trains suffer from insufficient real-vehicle bearing fault data, resulting in unsatisfactory diagnostic accuracy.

[0085] Existing technologies use experimental benches to generate bearing fault data for various operating conditions. However, the source domain bearing fault data differs significantly from the target domain bearing data and cannot cover more operating conditions. Some abnormal source domain bearing fault data reduces the accuracy of bearing fault diagnosis.

[0086] To overcome the aforementioned technical problems, this application provides a high-speed bearing fault diagnosis method based on a coupled model and adversarial network, comprising: obtaining a source domain and a target domain through dynamic simulation based on an axle box-vehicle-track coupled dynamic model; the source domain being a labeled bearing fault dataset with multiple operating conditions, and the target domain being an unlabeled bearing dataset; preprocessing the source and target domains to obtain a preprocessed source domain and a preprocessed target domain; pre-training a pre-constructed domain adversarial neural network based on the preprocessed source domain until the maximum number of iterations is reached to obtain a pre-trained domain adversarial neural network; and inputting the preprocessed target domain into the pre-trained domain adversarial neural network for adjustment to identify the bearing health status. Compared with the prior art, the source and target domains obtained by dynamic simulation in this application cover bearing faults under more operating conditions, reducing the difference between the source and target domains and reducing the data requirements of the domain adversarial neural network for source domain training data; and the parameters for improving domain discrimination in the pre-trained domain adversarial neural network include source domain sample weights, reducing the interference of abnormal bearing fault data in the source domain, while extracting bearing fault features at multiple scales through multi-scale dilated convolutional layers, thereby improving the accuracy of bearing fault diagnosis.

[0087] The technical solution of this application will be illustrated below through specific embodiments.

[0088] Firstly, such as Figure 1 As shown, this application provides a high-speed bearing fault diagnosis method based on a coupled model and adversarial network, including:

[0089] S1. Based on the coupled dynamic model of axle box-vehicle-track, dynamic simulation is performed to obtain the source domain and the target domain.

[0090] In this embodiment, the source domain is a labeled bearing fault dataset with multiple operating conditions. Labeled source domain samples are generated through dynamic simulation based on an axle box-vehicle-track coupled dynamic model. By updating parameter settings and amplitude scaling, source domain bearing fault datasets for various simulated operating conditions can be obtained, broadening the range of bearing operating scenarios and providing more training data for learning generalized features of bearing fault characteristics. By combining dynamic simulation data with experimental data, the generalization ability of the dynamic model is improved, helping to alleviate diagnostic limitations caused by insufficient real-vehicle fault data. Furthermore, in many transfer learning studies, experiments are often limited to the same equipment under constant or variable speed conditions. This application transfers constant-speed learning to different equipment under variable speed conditions. The target domain is unlabeled bearing data obtained experimentally, used to simulate fault data occurring in real vehicles, which is closer to the actual train operating environment and improves the reliability of bearing fault diagnosis.

[0091] In one embodiment, to more comprehensively and realistically simulate the train operation environment, based on the vehicle-track coupled dynamics model considering the vehicle's operating state and track structure parameters, an axle box bearing structural parameter-integrated axle box-vehicle-track coupled dynamics model is established. This model can reveal the vibration response when the axle box bearing experiences a localized fault. The overall modeling process for the axle box-vehicle-track coupled dynamics model is as follows: Figure 5-6 As shown, taking a certain type of high-speed train in China as the research object, a vehicle-track coupled dynamic model considering the axle box is constructed using the overall approach of co-simulation. First, solid models of the inner and outer rings, rolling elements, and cage components of the double-row tapered roller bearing are created in 3D modeling software (such as SolidWorks). Then, the solid models are imported into multibody dynamics simulation software (such as RecurDyn) to construct an axle box-vehicle-track coupled dynamic model for dynamic simulation of bearing signals, thereby generating a source domain bearing fault dataset for dynamic simulation.

[0092] In this embodiment, the bearing fault dataset in the source domain is: The target domain bearing dataset is in, This refers to the bearing fault data of the i-th source domain, which is labeled with the actual operating condition category. for Corresponding real-world working condition category; ns n represents the number of samples in the source domain bearing fault dataset labeled with real-world operating condition categories. t The number of samples in the target domain bearing dataset that are not labeled with the actual operating condition category; This refers to the bearing data for the i-th target domain that is not labeled with the actual operating condition category.

[0093] S2 performs data preprocessing on the source and target domains to obtain the preprocessed source and target domains.

[0094] In this embodiment, a sliding window method is used to segment the acquired source domain signal and target domain signal into samples of uniform length. During cross-device transmission, the source domain data acquired under variable speed conditions is resampled in the angle domain. The resampling process converts non-stationary signals into stationary signals, obtaining preprocessed source and target domains, thus improving the consistency and quality of the analyzed data.

[0095] S3, based on the pre-processed source domain, pre-train the constructed domain adversarial neural network until the maximum number of iterations is reached, and obtain the pre-trained domain adversarial neural network.

[0096] In this embodiment, the pre-trained domain adversarial neural network is pre-trained based on the pre-processed source domain until the maximum number of iterations is reached, resulting in a pre-trained domain adversarial neural network. The parameters for improving domain discrimination in the domain adversarial neural network include source domain sample weights, which are used to reduce the interference of abnormal bearing data in the source domain bearing fault dataset. Furthermore, the edge distribution and conditional distribution of the source domain and the target domain are aligned by jointly using the maximum mean difference loss, thereby improving the bearing fault diagnosis accuracy of the pre-trained domain adversarial neural network.

[0097] In one embodiment, such as Figure 2As shown, the domain adversarial neural network includes an input signal, a feature extractor, a neighborhood discriminator, a feature classifier, and a loss decision output. The feature extractor extracts bearing fault features from the source or target domain of the input signal using a multi-scale expansion method. The input of the neighborhood discriminator is connected to the output of the feature extractor to receive the bearing fault features. The parameters for improving the neighborhood discriminator are configured with source domain sample weights based on the difficulty of the neighborhood discriminator, and the differences between the bearing fault features corresponding to the source and target domains are obscured. Through feature input, a gradient inversion layer, and neighborhood discriminator, the bearing fault features are labeled with the source or target domain, thereby learning domain-invariant features to achieve [the desired outcome]. The system now features more robust generalization. The input of the feature classifier is connected to the output of the feature extractor to receive bearing fault features, perform label classification on the bearing fault features, and obtain the decision boundary. The loss determination output determines the classification loss based on the received label classification, and determines the domain discrimination loss based on the received domain labels. The joint maximum average difference loss is determined based on the feature extraction and feature classification of the source and target domains. The marginal distributions and conditional distributions between the source and target domains are jointly aligned, and the bearing health status of the bearing fault data is identified by minimizing the loss function value. The bearing health status labels include bearing normal label, bearing outer ring fault label, bearing roller fault label, and bearing inner ring fault label.

[0098] In one embodiment, such as Figure 3As shown, the domain adversarial neural network includes 5 convolutional layers, 1 multi-scale dilated convolutional layer, 1 adaptive max-pooling layer, and 2 fully connected layers. The feature extractor comprises 5 convolutional layers and 1 multi-scale dilated convolutional layer. Each convolutional layer consists of a convolution operation, batch normalization, and a ReLU activation function. The multi-scale dilated convolutional layer is located between the first and second convolutional layers. The second, third, fourth, and fifth convolutional layers are connected sequentially. The domain discriminator includes feature embedding, a gradient inversion layer, and domain discrimination. The feature classifier includes 2 fully connected layers. The first convolutional layer has 16 output channels, a kernel size of 15×15, a stride of 1×1, padding of 7×7, and an output size of (16, 1024). A multi-scale dilated convolutional layer also has 16 output channels and an output size of (16, 1024). The second convolutional layer has 32 output channels, a kernel size of 3×3, a stride of 1×1, padding of 2×2, and an output size of (32, 1024). The third convolutional layer has 64 output channels, a kernel size of 3×3, a stride of 1×1, and padding of 1×7. 1. The output size is (64, 512); the 4th convolutional layer has 128 output channels, a kernel size of 3×3, a stride of 1×1, and padding of 1×1, with an output size of (128, 512); the 5th convolutional layer has 256 output channels, a kernel size of 3×3, a stride of 1×1, and padding of 1×1, with an output size of (256, 256); the 1st fully connected layer is connected to the adaptive max pooling layer, with 256 output channels; the 2nd fully connected layer is connected to the 1st fully connected layer, and the output channels represent the number of bearing health state categories.

[0099] In one embodiment, such as Figure 4 As shown, step S3 pre-trains the constructed domain adversarial neural network based on the pre-processed source domain until the maximum number of iterations is reached, resulting in a pre-trained domain adversarial neural network, including:

[0100] S31, Initialize the network structure of the domain adversarial neural network, and determine the maximum number of iterations and the learning rate.

[0101] S32, based on the first parameter, second parameter, third parameter, feature classifier, feature extractor, source domain bearing fault dataset, number of samples in the source domain bearing fault dataset, number of health categories, and indicator function, determines the classification loss through the classification loss calculation formula.

[0102] S33, based on the preprocessed source domain, preprocessed target domain, the number of samples in the source domain bearing fault dataset, the number of samples in the target domain bearing dataset, the Gaussian kernel function, the feature classifier, and the feature extractor, the joint maximum average difference loss is determined by the joint maximum average difference loss calculation formula.

[0103] S34. Based on the preprocessed source domain, preprocessed target domain, number of samples in the source domain bearing fault dataset, number of samples in the target domain bearing dataset, source domain sample weights, feature classifier, feature extractor, and domain discriminator, the domain discrimination loss based on the source domain sample weights is determined by the domain discrimination loss calculation formula.

[0104] S35. Based on classification loss, domain discrimination loss, regularization coefficient, joint maximum average difference loss, and trade-off parameters, the loss function value is determined through the loss function calculation formula.

[0105] S36 updates the first, second, and third parameters respectively using adaptive motion estimation and learning rate, and performs iterative iterations until the maximum number of iterations is reached, thus obtaining a pre-trained neural network.

[0106] In this embodiment, to reduce the impact of outlier samples in the source domain, a weight allocation mechanism is introduced into the loss function, assigning weights to each source domain sample based on the evaluation results of the domain discriminator. A domain alignment strategy based on the joint maximum mean difference loss term is adopted to align the edge and conditional distributions of the source and target domains, thereby enhancing the generalization ability of feature extraction across different distribution domains. Specifically, batch normalization adjusts the output of the convolutional layer to a distribution with a mean of 0 and a variance of 1, accelerating the convergence speed during training. The ReLU activation function enhances the nonlinear expressive power of the network and improves the efficiency of extracting bearing fault features. The maximum number of iterations is greater than or equal to 500, for example, 300–500 iterations.

[0107] In one embodiment, the loss function is calculated as follows:

[0108]

[0109] Where, L(θ) f ,θ c ,θ d ) represents the loss function value; θ f This is the first parameter of the feature extractor;

[0110] θ c θ is the second parameter of the feature classifier. d This is the third parameter of the domain discriminator;

[0111] L c The loss is the classification loss; λ is the regularization coefficient. Loss is determined by the domain;

[0112] μ is the trade-off parameter; L JMMD This represents the combined maximum average difference loss.

[0113] In one embodiment, the classification loss is calculated as follows:

[0114]

[0115] Among them, L c For classification loss; θ f This is the first parameter of the feature extractor;

[0116] θ c This is the second parameter of the feature classifier;

[0117] n s The number of samples in the source domain;

[0118] K s The number of health categories; I[*] is the indicator function;

[0119] i is the source domain sample number; This represents the i-th bearing fault data in the source domain;

[0120] for Corresponding real-world operating condition categories;

[0121] For feature classifier G c right The k-th predicted working condition category is output.

[0122] For feature extractor G f right Output bearing fault characteristics.

[0123] In one embodiment, to suppress interference from anomalous samples on source domain samples, source domain sample weights are assigned based on the discrimination difficulty of the domain discriminator to train a robust domain adversarial network. The domain discrimination loss is calculated as follows:

[0124]

[0125] in, For domain discrimination loss; θ f This is the first parameter of the feature extractor;

[0126] θ d This is the third parameter of the domain discriminator;

[0127] n s n represents the number of samples in the source domain. t The number of samples in the target domain;

[0128] i represents the source domain sample number; j represents the target domain sample number;

[0129] Let be the source domain sample weight of the i-th sample in the source domain;

[0130] This represents the i-th bearing fault data in the source domain; This refers to the i-th bearing data in the target domain.

[0131] For feature extractor G f right Output bearing fault characteristics;

[0132] For the domain discriminator G d right The output prediction domain category;

[0133] For feature classifier G c right Output predicted operating condition categories;

[0134] For the domain discriminator G d right The output prediction domain category;

[0135]

[0136] In one embodiment, the joint maximum average difference loss achieves distribution alignment between domain categories by calculating the difference between the joint distributions of the source and target domains. The formula for calculating the joint maximum average difference loss is:

[0137]

[0138] Among them, L JMMD (θ f ,θ c ) represents the joint maximum average difference loss;

[0139] n s n represents the number of samples in the source domain. t The number of samples in the target domain;

[0140] i represents the source domain sample number; j represents the target domain sample number;

[0141] g(·) is the Gaussian kernel function;

[0142]

[0143] This represents the i-th bearing fault data in the source domain; This refers to the i-th bearing data in the target domain.

[0144] Gc For feature classifiers; G f For feature extractors.

[0145] In one embodiment, an adaptive motion estimation algorithm is used to solve and update the first, second, and third parameters. To prevent gradient vanishing or exploding during training, a residual block training strategy is employed. The parameter update process can be represented as follows:

[0146]

[0147] Where, β f θ is the first parameter of the feature extractor. c This is the second parameter of the feature classifier;

[0148] θ d L is the third parameter of the neighborhood discriminator. c For classification loss; Loss is determined by the domain;

[0149] L JMMD The maximum average difference loss is denoted by λ; λ is the regularization coefficient; μ is the tradeoff parameter.

[0150] η is the learning rate.

[0151] S4. The preprocessed target domain is input into the pre-trained domain adversarial neural network for adjustment in order to identify the bearing health status.

[0152] In one embodiment, during the pre-training phase of the domain adversarial neural network (DAN), a labeled bearing fault dataset from the source domain is used as input to train the constructed DAN. The source domain data first extracts features through a shared feature extractor, then feeds it into a label classifier to predict the fault category, and calculates the classification loss based on the true labels from the source domain. Simultaneously, the source domain data and unlabeled data from the target domain are input into a domain discriminator. An adversarial mechanism is constructed through a gradient reversal layer, causing the feature extractor to generate domain-independent features that cannot be accurately distinguished by the domain discriminator during optimization, thereby minimizing the feature distribution difference between the source and target domains. During training, the weighted sum of the classification loss and the domain adversarial loss is continuously optimized until the maximum number of iterations is reached, ultimately resulting in a pre-trained DAN with good generalization ability.

[0153] After pre-training, the fine-tuning phase begins. Unlabeled bearing data from the target domain is input into the pre-trained domain adversarial neural network (DAN). The DAN is then fine-tuned through continued adversarial training, further optimizing its parameters until the bearing health status output by the DAN is greater than or equal to a certain accuracy threshold (e.g., 90% or higher). Since the source and target domains use the same network structure and share weights during feature extraction, the fine-tuning process only requires further optimization of the existing DAN parameters, without retraining the network structure. This fine-tuning phase further enhances the DAN's adaptability to target domain features.

[0154] Finally, a fine-tuned domain adversarial neural network is used to identify faults in the unlabeled bearing data of the target domain. After being processed by a shared feature extractor, the target domain data is input into a label classifier for category prediction. The domain adversarial neural network outputs the fault type corresponding to each target domain sample, thus completing the fault diagnosis task for the unlabeled bearing data of the target domain.

[0155] In one embodiment, such as Figure 5 , 6 As shown, the source and target domains are preprocessed, and then input into a pre-trained domain adversarial neural network to obtain... Figure 7 Identified bearing health status diagram Figure 8 The confusion matrix diagram of the recognition results and Figure 9 t-SNE clustering diagram.

[0156] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0157] In a second aspect, embodiments of this application provide an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the method as described in any one of the first aspects above.

[0158] Thirdly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method as described in any one of the first aspects above.

[0159] Fourthly, embodiments of this application provide a computer program product that, when run on an electronic device, causes the electronic device to perform the method described in any one of the first aspects above.

[0160] It is understood that the beneficial effects of the second to fourth aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here.

[0161] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms.

[0162] The computer-readable medium may include at least: any entity or device capable of carrying computer program code to a photographic device / terminal device, recording media, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media. Examples include USB flash drives, portable hard drives, magnetic disks, or optical discs. In some jurisdictions, according to legislation and patent practice, computer-readable media may not be electrical carrier signals or telecommunication signals.

[0163] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0164] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0165] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0166] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0167] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A high-speed bearing fault diagnosis method based on a coupling model and a generative adversarial network, characterized in that, The application relates to a bearing health state recognition method based on a domain adversarial neural network. The method comprises the following steps: a source domain is obtained based on a shaft box-vehicle-track coupling dynamics model for dynamics simulation, and a target domain is obtained, the source domain is a plurality of working condition bearing fault data sets marked with labels, and the target domain is an unmarked bearing data set; data preprocessing is performed on the source domain and the target domain to obtain a preprocessed source domain and a preprocessed target domain; a domain adversarial neural network is pre-trained based on the preprocessed source domain until a maximum iteration number is reached, and a pre-trained domain adversarial neural network is obtained; the preprocessed target domain is input into the pre-trained domain adversarial neural network for adjustment to identify the bearing health state; wherein the parameters for improving domain discrimination in the domain adversarial neural network include source domain sample weights for reducing the interference of abnormal bearing data in the source domain; 2. The high-speed bearing fault diagnosis method based on the coupling model and the adversarial network according to claim 1, wherein, the convolutional layer of the domain adversarial neural network comprises a multi-scale dilated convolutional layer for extracting bearing fault features of multiple scales. The domain adversarial neural network comprises an input signal, a feature extractor, a domain discriminator, a feature classifier and a loss determination output; wherein the feature extractor is used to extract bearing fault features from the preprocessed source domain or the preprocessed target domain of the input signal in a multi-scale dilated manner; the input end of the domain discriminator is connected to the output end of the feature extractor, and is used to receive the bearing fault features, set the source domain sample weights according to the difficulty of the domain discriminator in the parameters for improving domain discrimination, and confuse the difference between the bearing fault features corresponding to the source domain and the target domain, and the domain labels of the bearing fault features are marked by the feature input, the gradient reversal layer and the domain discrimination; the input end of the feature classifier is connected to the output end of the feature extractor, and is used to receive the bearing fault features and classify the labels of the bearing fault features; 3. The high-speed bearing fault diagnosis method based on the coupling model and the adversarial network according to claim 2, wherein, the loss determination output receives the classification loss determined by the label classification, receives the domain discrimination loss determined by the domain label, determines the joint maximum mean difference loss according to the feature extraction and the feature classification of the source domain and the target domain, aligns the marginal distribution and the conditional distribution between the source domain and the target domain, and identifies the bearing health state of the bearing fault data by minimizing the loss function value. The domain adversarial neural network comprises five convolutional layers, one multi-scale dilated convolutional layer, one adaptive maximum pooling layer and two fully connected layers; 4. The high-speed bearing fault diagnosis method based on the coupling model and the adversarial network according to claim 1, wherein, wherein the feature extractor comprises five convolutional layers and one multi-scale dilated convolutional layer, each convolutional layer is composed of a convolution operation, batch normalization and a ReLU activation function, the one multi-scale dilated convolutional layer is between the first convolutional layer and the second convolutional layer, and the second convolutional layer, the third convolutional layer, the fourth convolutional layer and the fifth convolutional layer are sequentially connected; the domain discriminator comprises feature embedding, a gradient reversal layer and domain discrimination; and the feature classifier comprises two fully connected layers. The domain adversarial neural network is pre-trained based on the preprocessed source domain until a maximum iteration number is reached, and a pre-trained domain adversarial neural network is obtained, which comprises the following steps: the network structure of the domain adversarial neural network is initialized, the maximum iteration number and the learning rate are determined; the classification loss is determined by a classification loss calculation formula based on the first parameter, the second parameter, the third parameter, the feature classifier, the feature extractor, the preprocessed source domain, the sample number of the source domain, the health class number and the indicator function; The joint maximum mean difference loss is determined by a joint maximum mean difference loss calculation formula based on the pretreated source domain, the pretreated target domain, the sample quantity of the source domain, the sample quantity of the target domain, a Gaussian kernel function, a feature classifier and a feature extractor; The domain discrimination loss based on the source domain sample weight is determined by a domain discrimination loss calculation formula based on the pretreated source domain, the pretreated target domain, the sample quantity of the source domain, the sample quantity of the target domain, the source domain sample weight, a feature classifier, a feature extractor and a domain discriminator; The loss function value is determined by a loss function calculation formula based on the classification loss, the domain discrimination loss, a regularization coefficient, the joint maximum mean difference loss and a trade-off parameter; The first parameter, the second parameter and the third parameter are updated respectively by an adaptive motion estimation mode and a learning rate, and a loop iteration is performed until the iteration number reaches a maximum iteration number, so as to obtain the pre-trained neural network.

5. The high-speed bearing fault diagnosis method based on the coupling model and the adversarial network according to claim 4, wherein, The calculation formula of the loss function is: wherein L(θ f ,θ c ,θ d ) is a loss function value; θ f is a first parameter of the feature extractor; θ c is a second parameter for the feature classifier; θ d is a third parameter for the domain discriminator; L c is the classification loss; λ is the regularization coefficient; is the domain discrimination loss; μ is a trade-off parameter; L JMMD is the joint maximum mean discrepancy loss.

6. The high-speed bearing fault diagnosis method based on the coupling model and the adversarial network according to claim 4, wherein, The calculation formula of the classification loss is: wherein L c is a classification loss; θ f is a first parameter of the feature extractor; θ c is a second parameter of the feature classifier; n s is the number of samples in the source domain; K s n is the number of health categories; I[*] is an indicator function; i is the source domain sample number; is the i-th bearing data of the source domain; To corresponding real-world class; G is a feature classifier c to the kth predicted operating condition class output; To the feature extractor G f To Output bearing fault features.

7. The high-speed bearing fault diagnosis method based on coupling model and adversarial network according to claim 4, characterized in that, The calculation formula of the domain discrimination loss is: wherein, is a domain discrimination loss; θ f is a first parameter of the feature extractor; θ d is a third parameter for the domain discriminator; n s is the number of samples of the source domain; n t is the number of samples of the target domain; i is a source domain sample number; j is a target domain sample number; is the source domain sample weight for the i-th sample in the source domain; xi for the i-th bearing fault data of the source domain; xi for the i-th bearing data of the target domain; G is a feature extractor f to outputted bearing fault features; for the domain discriminator G d to the predicted domain class; To feature classifier G c To Outputted predicted operating condition class; for the domain discriminator G d to the output predicted domain class; 8. The high-speed bearing fault diagnosis method based on coupling model and adversarial network according to claim 4, characterized in that, The calculation formula of the joint maximum mean difference loss is: wherein L JMMD (θ f ,θ c ) is the joint maximum mean discrepancy loss; n s is the number of samples of the source domain; n t is the number of samples of the target domain; i is a source domain sample number; j is a target domain sample number; g(·) is a Gaussian kernel function; xi as the i-th bearing fault data of the source domain; xi as the i-th bearing data of the target domain; G c is a feature classifier; G f is a feature extractor.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, The processor executes the computer program to realize the high-speed bearing fault diagnosis method based on the coupled model and the adversarial network according to any one of claims 1 to 8.

10. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1-9. The computer program is executed by the processor to realize the high-speed bearing fault diagnosis method based on the coupled model and the adversarial network according to any one of claims 1 to 8.

Images